Skip to content

CSV, CRM, and sample data

What it represents

There are three ways to have data in ax1om, and they are not the same thing.

Sample data is synthetic and provisioned for you. It exists so you can walk the whole product before deciding anything. It is not your CRM and it is labeled that way everywhere it appears.

A CSV is your data with no access required. Export leads or contacts, drop the file in, and train on it.

A CRM connection is the full path. Connect Salesforce or HubSpot and ax1om reads your objects and fields directly.

How it’s calculated

Sample data is already seeded, so scores, models, runs, factors, and performance views all work immediately, on records that are not real.

A CSV upload parses the file, detects column types, and stands the columns up as a data source you can build a score on. There is no CRM authorization step and no admin conversation to have first.

Connecting a CRM runs schema discovery: ax1om pulls your objects, fields, record counts, and how the data is distributed across years, then tiers fields by how populated they are so you can see what is worth training on. Sample data and CSV sources do not count against your plan’s connection limit.

The walk, whichever source you started from

The path from data to a first model is the same for all three. Only the records underneath it differ.

  1. Open Scores and see what is already there

    Active scores sit at the top. Anything still a draft, paused, or provisioned as sample data sits below them, and sample entries carry a Sample label everywhere they appear so a demo score is never mistaken for a live one.

    The Scores list, headed 2 active scores across your CRM. An Active group holds Lead Conversion 2, tagged Salesforce, and In-Market Timing. A Drafts and inactive group below it holds three entries, each tagged Sample.
    The Scores list · the Sample tag is the sample-data org, not your CRM.
  2. Create a score by picking what you want to predict

    Templates are grouped by what they score: people, accounts, or deals. Each one is a question rather than a configuration, and picking it seeds the entity, the success criteria, and a sensible starting field set. Templates for features that have not shipped are shown but labelled, and there is a start-from-scratch option at the bottom for anything the list does not cover.

    The Create a Score template picker, subtitled Pick what you want to predict. We'll help you build a model to power it. A People group offers Lead Conversion. An Accounts group offers Account Fit and Expansion Score. A Deals group offers Win Likelihood. Two further templates are dimmed and labelled Coming soon, and a start from scratch option sits at the bottom.
    Templates are questions, not settings · the wording under each one is the question the score answers.
  3. Work the three wizard steps in order

    Data, then engagement, then review and train. The first step is where the source you brought actually matters, and it is the step worth slowing down on: everything after it is reading what you chose here.

    The model wizard step indicator: step 1 Data is current, followed by 2 Engagement and 3 Review and Train.
  4. Pick fields from the available list

    Fields are grouped by the object they came from, with the count beside each group name and a search box over the whole set. The bar and percentage on each row is how often that field is populated, so an empty column is visible before you spend a training run on it. Suggest fills a starting set for you, and the star marks a field as recommended.

    The Available fields panel, subtitled Click to add, with a Suggest button and a field search box. An Account group of 20 fields is expanded, showing a Recommended band. Billing Country is selected and 100% populated. Account Type, Billing City, Billing State/Province, and Billing Zip/Postal Code follow at 0%, each showing its CRM API name underneath and a linked tag.
    The Available fields panel · the percentage is fill rate, and 0% means the column is there but empty.
  5. Read the health panel as you go, not afterwards

    Every field you add lands here with its type, its fill rate, and a badge for how many distinct values it holds. The footer counts what you have selected against the merged record total. This is the cheapest moment in the whole process to notice that half your selections are empty.

    The Selected Fields and Health panel showing 7 fields selected, grouped as Account with 2 fields and Lead with 5. Billing Country is 100% populated; Account Type is 0%; Title, Company, and Lead Source are each 40%, with cardinality badges reading Med, High, and Low. The footer reads 7 fields · 12,000 merged records.
    Selected fields and health · fill rate on the left, how many distinct values on the right.

What it means for you

The difference is what happens after the first model. Scheduled refresh needs something to refresh from, and writeback needs somewhere to write. Both of those want a live connection. On sample data you can see how scores behave; you cannot push them anywhere, because there is no CRM behind them. A CSV is a snapshot, so a new week of data means a new upload.

Worked example: you spend an hour on sample data, train a model, and read its AUC, feature analysis, and per-record factors to decide whether the approach holds up. Then you upload a 30,000-row export of last year’s leads and train on your own history · that is where the number starts meaning something about your market. When it does, connect the CRM, set a weekly cadence, point the score at a field, and the whole thing runs without you.

Any of the three is a legitimate place to start. Only the last one is a place to stay.